delta layers

Delta clips are authored relative to ‘CFrame.identity’. Identity means no effect, translation and rotation are an offset applied to the current local transform.

local delta_state = {
  binding = animation.bind_animation(recoil_asset, rig.desc),
  time_position = 0,
  weight = 1,
}

local prepared_deltas = solver.prepare_delta_tracks({
  {
    asset = recoil_asset,
    state = delta_state,
    mode = "additive",
  },
})

The frame order is always:

solver.solve_prepared(rig, native_prepared)
solver.apply_delta_prepared(rig, prepared_deltas)

Danger

apply_delta_prepared() edits rig.transforms in place. Calling it again without a fresh native solve compounds the previous delta result.

Modes and order

For each bound target, Reel first applies the delta weight from identity. It then chooses the operation:

additive:    base * weightedDelta
subtractive: base * inverse(weightedDelta)

Delta preparation preserves the array order you supply. It does not sort by priority. Rotation composition is not commutative, so layer order is observable:

base * aim * inverse(correction) * recoil

An immediately adjacent additive and subtractive use of the same weighted delta cancels exactly.

Authoring notes

• Author a neutral channel as CFrame.identity
• Add an identity key at time zero when the layer should begin neutral
• Missing channels leave that rig target unchanged, allowing partial-body overlays
• Keep the binding paired with the same asset that created it

The animation loader duplicates a channel’s first key at time zero when that key starts later. For delta clips, an omitted identity key can therefore make a non-identity offset active immediately.

Delta tracks do not use native blend scratch. Their cost is proportional to their bound channels.